Social discounting and risk attitudes, time preferences and social
preferences: An experimental study
Social discounting and risk attitudes, time preferences and social
preferences: An experimental study
by
Celeste Campher
Student number: 1996632822
Submitted in fulfilment of the requirements in respect of the Doctoral degree
qualification Doctor Philosophiae (PhD) Economics in the Department of
Economics and Finance in the Faculty of Economic and Management Sciences at
the University of the Free State.
Submitted: 30 June 2020
Promoter: Professor Frederik Booysen
School of Economics and Finance
University of the Witwatersrand
Co- Promoter: Dr. Sevias Guvuriro
Department of Economics and Finance
Dedication
I
DECLARATION
I, Celeste Campher, declare the following:
I.
The Doctoral Degree research thesis that I herewith submit for the
Doctoral Degree qualification Philosophiae Doctor (PhD) Economics
at the University of the Free State is my independent work, and that I
have not previously submitted it for a qualification at another institution
of higher education,
II.
I am aware that the copyright is vested in the University of the Free
State,
III. All royalties as regards intellectual property that was developed during
the course of and/or in connection with the study at the University of
the Free State, will accrue to the University.
II
ABSTRACT
Altruism, the principle or practice of concern for the welfare of others, is a key element of human behaviour. It is vital to gain more knowledge on how altruism is associated with economic and other social preferences in order to get a more nuanced understanding of peoples’ economic and social interactions. This thesis aims to extend research on social discounting in two areas, namely: how preferences for giving and social discounting differ in laboratory and field subjects, and how preferences for giving and social discounting are associated with various other economic and social preferences. The study comprises a series of conventional laboratory experiments and a set of artefactual field experiments with students and staff, respectively, from the University of the Free State in South Africa. These experiments employ social discounting tasks (SDT), multiple price lists (MPLs) for eliciting risk and time preferences, and ultimatum games (UG) and a trust games (TG) as tools for preference elicitation. Descriptive statistical analyses that investigate associations between key variables and appropriate regression models are employed in the thesis. A number of key findings are highlighted in the thesis. First, the thesis finds no consistent evidence of significant differences in preferences for giving and social discounting across laboratory and field subjects. Secondly, the thesis finds that the motivations for altruistic behaviour amongst laboratory and field subjects differ significantly with regard to the choice of recipients and the nature of the relationships with the recipients. The third finding suggests that risk and time preferences are correlated with altruism, as measured by giving and social discounting, but that this association is complex and non-linear in nature. Finally, the thesis provides evidence that altruism is associated with egalitarianism, reciprocity and, to a lesser extent, with trust, as well as with the trustworthiness. Giving and social discounting, therefore, are important features of human behaviour and requires further investigation pertaining to its economic and social consequences.
III
ACKNOWLEDGEMENTS
First and foremost, my almighty and gracious God. This thesis would not have materialized
had it not been for Your ever-present mercy and grace in my life. You have blessed me with an immeasurable amount of strength, courage and persistence during the course of completing this document. I cannot begin to describe how grateful I am for how You never left my side during this entire journey to the point where you actually carried me during the last year. I am not worthy of your grace and mercy Lord, but I am eternally grateful. All the glory and honour goes to You. Thank You Lord.
To Cuan, my husband, and our four beautiful children, Matthew, Kirsten, Danni and Brody. Thank you for your never ending love, support and patience during the last four years. Thank you for the sacrifices you made for me so that I could pursue my dream, it has not gone unnoticed. Thank you also for putting up with my moods. I know this journey has been difficult for all of us but I am so grateful that I could share it with all of you. To my children, I hope I have inspired you to always work towards achieving your goals and fulfilling your dreams. I have no doubt that each one of you are capable of achieving whatever you put your minds to and I pray you all grow into your full potential. I love you all more than words could ever say.
My sincerest and deepest gratitude to my promoters, Professor Frederik Booysen and Dr. Sevias Guvuriro. I know at times it has been a frustrating journey for you as well but I thank you for your patience with me. You have endowed me with a wealth of knowledge and your tutelage has provided me with a sound foundation to pursue my career as a Behavioural Economist. You both have inspired me to be a better academic and researcher, and I will carry the lessons learned from you during the process of completing this thesis for the remainder of my academic career. Thank you for everything you did to help me complete this thesis. I will never forget it.
To my parents, Edward and Blanche Tait. Mummy and Daddy, thank you for always believing in me and for encouraging me to be studious and to pursue this dream. As ex-teachers yourselves I know I get my passion for academics and my love for studying from both of you as well as my work ethic and perseverance. Thank you for setting such a good example to our family. Thank you for your constant words of encouragement and prayers, and for always being
IV
willing to help out with looking after my children when I needed to put in the hours to complete this thesis. I love you with all my heart.
To my siblings, David and Anthea. Thank you for being in my corner and for your silent prayers. You both inspire me to be a better person, parent, daughter and sibling on a daily basis. I have now completed the next step in my academic journey and I truly hope I have encouraged you to do the same, whether it is in your studies or careers. I love you and appreciate you both so much.
Finally, to my colleagues in the Department of Economics and Finance at the University of the Free State. Thank you for your unwavering support and kindness and for your patience with my frustrations and my moods. I am truly fortunate to be able to work with such an amazing group of people.
I am very grateful that I was able to complete this thesis with the financial support from the National Research Foundation’s (NRF) Human and Social Dynamics Research Programme and the Thuthuka Programme.
V
TABLE OF CONTENTS
Declaration………..i
Abstract………...ii
Acknowledgements……….iii
Chapter 1: General Introduction………..1
1.1 Background……….1
1.2 The social discounting task: Giving and social discounting...………1
1.3 Rationale for the study………5
1.4 Aim and objectives……….6
1.5 Methodology………...7
1.6 Organisation of the thesis………....9
Chapter 2: Paper 1………..10
Chapter 3: Paper 2………..46
Chapter 4: Paper 3………..86
Chapter 5: General conclusion………126
References……….130
1
CHAPTER ONE: GENERAL INTRODUCTION
1.1 Background
People often encounter the reality of making decisions that affect their own welfare and/or the welfare of other people. When such decisions are made, individuals’ preferences play a significant role. At the same time, the subjective closeness or how socially distant the other person involved is to the decision maker, i.e. social distance (Trope & Liberman, 2003), is a determining factor of the willingness of the decision maker to sacrifice his or her own welfare. This reality qualifies the importance of altruism or selflessness, i.e. the principle or practice of concern for the welfare of others, in helping us understand or explain human behaviour. In addition, how altruism associates with other preferences (i.e. risk, time and other social preferences traits) is vital knowledge to build.
This general introduction to the thesis presents a brief description of the experimental measure of altruism adopted in this thesis, the problem statement, aim and objectives, as well as an outline of the methodological framework of the thesis. The chapter concludes with a brief summary of the organisation of the thesis.
1.2 The social discounting task: Giving and social discounting
The economic definition of altruism is described by Fehr and Fischbacher (2003) as “costly acts that confer economic benefits on other individuals”. This definition is consistent with the biological definition where Krebs and Davies (1993) refer to altruism as “acting to increase another individual’s lifetime number of offspring at a cost to one’s own survival and reproduction”. In this thesis, altruism is measured by giving and social discounting. According to Rachlin and Jones (2008), giving is measured as the amount of money a participant is willing to forego to give a fixed amount of money to another person at a specific social distance. Rachlin and Raineri (1992) first introduced the concept of social discounting as an individual’s willingness to forego an outcome for him/herself in exchange for a larger outcome for someone else. Simon (1995) incorporated the concept of altruism into utility functions and suggested that a person’s allocation of available goods can be described as (1) current consumption by the actual person, (2) consumption by the same person at later times [delay discounting], and (3) consumption by other people [social discounting].
2
Other early research studies using experimental economic games to explore altruism and other forms of human behaviour include, amongst others, Fehr and Gächter (2002), Frank, Gilovich, and Regan (1993), and Milinski, Semmann, and Krambeck (2002b). A common experimental tool used for this purpose is the dictator game.
In a perfectly standard dictator game the two players are to share a sum of money given to them by the experimenter (e.g. $10) between themselves but only the dictator (proposer) can determine the size of the share. According to standard economic theory- based on assumptions of rationality and self-interest- a dictator should keep 100% of the endowment and give nothing to the recipient. However, this is not what is shown by the literature. For instance, Guala and Mittone (2010) show that only 40% of subjects playing the role of the dictator keep all the money to themselves, while the majority of participants give on average 20% of their endowment to the recipients. Engel’s (2011) meta- analysis study reviews 131 research papers where the DG was employed with 616 different treatment effects and finds that dictators give on average 28.35% of their endowment pie to the recipients. In only 6 of the 616 treatments, dictators gave on average zero. The actions taken by the dictator in the DG may be indicative his/her preferences for altruism and fairness. A dictator who is willing to give of their endowment for the recipient is considered to be altruistic, while those who are concerned with an equitable outcome can be considered to be behaving fairly. In its standard form, the DG provides researchers with a simple and effective tool to observe and study these two real human behaviour phenomena. Over the years, researchers have used various versions of the standard dictator game such as one-shot games versus repeated games. A number of different treatment effects are applied to the standard dictator games in Engel’s (2011) meta-analysis study. When manipulating for social distance, Engel’s meta-study found a surprising negative effect with dictator generosity declining for recipients at close social distances. Empirically, researchers have manipulated social distance by experimentally inducing differences in the degree of anonymity between dictator and recipient. Examples of such experiments include Bechler et
al. (2015), Bohnet et al. (1999), Brañas-Garza et al. (2010), Charness and Gneezy (2008),
Goeree et al. (2010), Leider et al. (2009), and Margittai et al. (2015).
Jones and Rachlin (2006) measure social discounting as a participant’s willingness to give a fixed amount (for example, €75 in figure 1 below) to another person, using a social discounting task (SDT). The SDT has become a common instrument used in psychology and behavioural psychology experiments as a tool to measure altruism.
3
Figure 1: Jones and Rachlin’s (2009) Social Discounting Task
Source: Jones and Rachlin (2009:64)
In laboratory experiments that implement social discounting tasks (SDT), researchers have attempted to measure the relationship between altruism and social distance, and whether social distance can be used as a predictor in a person’s social interactions. In a typical SDT experiment (e.g. Jones & Rachlin, 2006; Olson et al., 2016; Locey & Rachlin, 2011), subjects are asked to imagine a list of people at a number of social distances, ranging from #1 closest to #100 furthest, and to choose between a declining amount of money for themselves and a fixed amount of money for the recipient at each social distance. An example of the task used by Jones and Rachlin (2009) to measure the relationship between social discounting and social distance is shown in Figure 1. In all the research papers on social discounting reviewed for this thesis, participants in these studies indicated that they would be much more generous to those close to them and much less generous to the socially distant. Such studies include, amongst others, Belisle et al. (2019), Booysen et al. (2018), Bradstreet et al. (2012), Ito et al. (2011), Jones and Rachlin (2006 & 2009), Leider et al. (2009), Locey et al. (2013), Locey and Rachlin (2013 & 2015), Margittai et al. (2015), Osiński (2009), Rachlin and Jones (2008), Soutschek et al. (2016), Strombach et al. (2013 & 2016), Yi et al. (2011 & 2016), Wu et al. (2019), and Ziegler
4
and Tunney (2012). Sharp et al. (2012) interpret such form of discounting to mean that most people assign more value to the welfare of close affiliates than they do to the welfare of distant affiliates. These studies validate Rachlin and Jones’s (2006) original result that rewards to others are discounted with increased social distance, and one can argue that the use of the social discounting task to measure associations between altruism and social distance is well documented.
Although the social discounting literature is relatively small, the consistency of the results across studies and now populations suggests that the social discounting task may be assessing something fundamental about the manner in which humans make choices about sharing within social networks. The SDT also allows the experimenter to gain a nuanced understanding of personal interactions in social relationships amongst the subject pool, especially where detailed information on recipient characteristics is collected, as is the case in this study. The experiment in this paper collects a variety of detail on the nature and quality of the relationship between the sender and recipient by using a questionnaire.
In this thesis three measures of altruism are estimated with the aid of data collected in the Social Discounting Task. The use of these multiple measures of altruism, i.e. crossover value, k’, and AUC allows for robustness checks in the actual analysis. A monetary value for the crossover value is estimated as the mean point at which a subject switched from choosing option A (selfish option) to option B (sharing option) for each social distance – and is captured as such in the task-level data set (i.e. for each social distance table). The social discounting functions determined from the crossover values, and which are used to estimate social discounting rates, the second measure of altruism employed in this study, is assumed to be hyperbolic (Jones & Rachlin, 2006), and expressed as follows:
where νij represents the value that person i attaches to the welfare of person j, Ai represents the
value person i associates with her own welfare, and Nij is the rank person i assigns to person j
among i’s full list of associated people. The constant k′ , one aggregate measure of altruism, measures the steepness of discounting and the greater k′ is, the greater the degree of social discounting and the lower the degree of altruism (Sharp et al., 2012). Another aggregate
5
measure of altruism, i.e. Area under the curve (AUC) is derived from each participant’s series of crossover values. The AUC index is a normalised measure not dependent on functional form (Myerson et al., 2001) and is constructed for each subject by making use of the Excel solver toolkit of Reed, Kaplan and Brewer (2012). The value of this index varies from 1.0 (no discounting) to 0.0 (complete discounting) (Locey et al., 2011). The k′ and AUC are both measures of the same phenomenon, social discounting, however, since AUC has less discriminant ability than k′, the analysis will include the use of both variables.
1.3 Rationale for the study
There is increasing evidence that human behaviour and decision making are influenced by various preferences such as patience, risk tolerance, fairness, generosity and trust, to name but a few. Research that makes use of experimental games such as the SDT to elicit preferences and to investigate associations amongst preferences has made significant strides, particularly in the field of social discounting research. According to Tiokhin et al. (2019), there are over 50 published research papers in the last decade that have documented findings on the social discounting phenomenon in laboratory or field settings. The standard social discounting protocol, namely the social discounting task, has been used in its standard form to assess altruism (Rachlin & Jones, 2006), as well as in modified versions to measure the relationship between social discounting and time discounting, and probability discounting (Bialaszek et al., 2019; Bickel et al., 2014; Jin et al., 2017; Jones & Rachlin, 2009; Osiński & Karbowski, 2017; Yi et al., 2011). However, there are still some research gaps that motivate the current study:
First, comparative social discounting studies employing the SDT as a measure of altruism
amongst both laboratory and field subjects are limited. Although previous studies on social discounting all confirm this behavioural bias, the majority of these studies were conducted with laboratory subjects only, namely university students. According to Tiokhin et al. (2019:3), “of 43 groups of participants from 21 publications, 40 groups were from the USA and/or university
students”. Laboratory subject pools represent a very limited sample of the overall population
and as such cannot be relied upon to infer assumptions about human behaviour on the whole.
Secondly, studies where economics preferences such as risk attitudes and time preferences, as
well as other social preferences are elicited alongside measures of altruism, with a focus on determining whether the latter is associated with the former preference sets are limited.
6
Previous research on social discounting and other economics and social preferences focuses mainly on the presence and extent of social discounting and its behavioural consequences, rather than attempting to explain the existence of these former mentioned preferences or links between social discounting and these preferences. Behavioural economics explores decision making and the motives underlying our decisions, as well as the preferences exhibited by our decisions. However, merely identifying and knowing about these preferences do not suffice for the further development of human behaviour theory. Since most preferences may be interrelated, research needs to investigate how other economic and social preferences are associated with altruism, measured here by estimates of giving and social discounting obtained with the aid of the social discounting task (SDT).
1.4 Aim and objectives
This thesis aims to extend the research on altruism in two areas, namely how preferences for giving and social discounting differ in laboratory and field subjects and how preferences for giving and social discounting are associated with various other economic and social preferences.
The main objectives of the study are:
(i) To elicit preferences for giving and social discounting in a group of laboratory and field subjects in order to determine if [Paper 1]:
- preferences for giving and social discounting differ amongst laboratory and field subjects.
- the dynamics underlying giving by laboratory and field subjects play out differently in terms of the characteristics of the recipients selected at each social distance.
(ii) To elicit risk and time preferences amongst a group of laboratory and field subjects alongside the SDT in order to determine if [Paper 2]:
- risk aversion and patience are associated with giving.
- risk aversion and patience are associated with social discounting.
(iii) To implement the ultimatum (UG) and trust (TG) games alongside the SDT in a group of laboratory and field subjects in order to determine if [Paper 3]:
7
- preferences for giving and social discounting are associated with a set of social subject types.
1.5 Methodology
The discipline of economics has expanded from a field dominated by the use of uncontrolled field and market data to one which includes many modes of research to test theory and inform policy. The popularity of laboratory and field experiments in economics has grown, with each type of experiment generating its own set of advantages and disadvantages relative to the traditional reliance upon field and market data. Harrison and List (2004) define a conventional lab experiment as one that employs a standard set of subjects, an abstract framing, and an imposed set of rules. A common criticism amongst researchers of laboratory experiments is on the reliance on the use of students as experimental subjects. The argument is that conclusions drawn from laboratory experiments with student subjects only may differ from the real-life context. An artefactual field experiment is the same as the latter, but with a non-standard set of subjects. This type of experiment allows the researcher to test the validity of results amongst a “mixed” group of subjects and allows for generalizability on theories of human behaviour. The current study consists of a series of conventional laboratory experiments in the case of student subjects and a set of artefactual field experiments with staff subjects from the University of the Free State.
The use of experimental games by psychologists and economists has gained much attention of late and has grown in attractiveness as a research tool for the analysis of human behaviour and decision-making, as well as for eliciting various preferences that form the basis for the underlying motives behind most decision-making processes. Economic experiments differ from psychological experiments in a number of ways. Economists make use of defined scripts, repeated trials, monetary payments based on performance, and avoid deception when performing experiments (Hertwig & Ortman, 2001). Researchers including Wilcox (1993) and Harrison (1994) believe that unless subjects are offered an incentive compatible payment schedule, their responses will not represent what they would do if they were given the task for real. According to Read (2005) the use of monetary incentives in performing experiments can influence the behaviour of subjects in three ways, namely, “cognitive exertion”, whereby subjects put more thought into their decisions; “motivational focus”, with the incentive causing the subjects to change their goals; and “emotional triggers”, whereby the incentive acts as a
8
trigger or prerequisite for a response. Furthermore, economists usually argue that financial rewards create a more realistic environment within the lab (Rosenboim & Shavit, 2012), causing subjects to consider their decisions more carefully (Carpenter et al., 2005). Psychologists, on the other hand, tend to believe that experimental subjects are generally intrinsically motivated and need no financial reward for decision-making (Camerer & Hogarth, 1999). Ben-Ner et al, (2008) found that the average subject behaves essentially the same as a dictator in an incentivized experiment and in a hypothetical experiment without money. They argued that the impact of incentives on the generosity of a dictator is much more complex and that researchers should consider the importance of individual characteristics in decision-making behaviour. However, Bühren and Kundt (2015) conducted an experimental study on social preferences using dictator games and found differences in social preferences of subjects who received low-stakes monetary incentives as compared to subjects who received hypothetical rewards. The experiment conducted at the University of Hamburg involved 80 subjects making incentivized decisions and 70 subjects making hypothetical choices in a number of DG experiments. The study found the incentivized subjects to be slightly more generous compared to the hypothetical subjects.
All of the three economic experiments conducted in this study comprised of a detailed list of instructions for each experimental tool, entailed the incentivisation of the tasks according to a payment procedure which randomly selected a monetary award per task for all participants, and practiced no deception of participants. Most studies employing social discounting experiments do not provide real incentives. This study attempts to do that by conducting a social discounting experiment within the guidelines for Economics experiments. According to Hertwig and Ortman (2001), economists run experiments to either (1) test decision – theoretic or game – theoretic models; (2) explore the impact of institutional details and procedures; or (3) improve understanding of policy problems. These experiments entail either constructing small-scale abstractions of real-world problems, translating theoretical models into laboratory set-ups, or providing experimental subjects with instructions that supply descriptions of players, their action choices, and possible pay-offs (Hertwig & Ortman, 2001). The latter methodological design is applied in the study, whereby the data for the thesis are collected from laboratory and field subjects by employing (1) a social discounting task (SDT); (2) multiple price lists for eliciting risk and time preferences; and (3) an ultimatum game (UG) and a trust game (TG). Paper 1 draws on the SDT data collected for the entire subject pool, while
9
preferences from the multiple price lists (MPL’s). Paper 3 is based on the data from the SDT, together with the data on the social subject types elicited from the UG and TG.
The methodology followed in this thesis is deductive by nature, whereby the hypotheses formulated in each of the three individual papers are empirically scrutinised using primary data collected from a laboratory and field subjects using various experimental tools. Descriptive statistical analyses that investigate associations between key variables and use Chi2, t-tests and
F-tests, and appropriate regression models are employed in the thesis.
1.6 Organisation of the thesis
The thesis is structured as follows: Chapter 1 focuses on the general introduction of the study. Chapters 2 to 4 comprise three individual research papers. Paper 1 entails a comparative study on the differences in altruistic behaviour of laboratory and field subjects and on the dynamics underlying the differences in altruistic behaviour in these two settings. Paper 2 empirically explores whether altruism is associated with risk and time preferences by employing an SDT to measure altruism, and separately eliciting risk and time preferences with a series of multiple price lists (MPLs). Paper 3 investigates associations between altruism, as measured by giving and social discounting, with social subject types uniquely identified in the study based on the strategy method responses of subjects in the UG and TG. Finally, Chapter 5 presents a brief integrated conclusion.
10
CHAPTER 2: PAPER 1
A comparative analysis of giving and social discounting amongst
laboratory and field subjects in South Africa
11 Table of Contents List of Tables……….12 List of Figures………...12 Abstract………...13 1. Introduction……….14 2. Experiment………..16 3. Measures……….18 4. Analysis……….20 5. Results………22 5.1 Subject characteristics………...22
5.2 Giving and social discounting by subject-type………..25
5.3 Subject-type and recipient characteristics……….33
6. Discussion………....42
12
List of Tables
Table 1: Sample descriptive statistics………24 Table 2A: Mean crossover values (Rand), by social distance and subject-type………...26 Table 2B: Median crossover values (Rand), by social distance and subject-type………27 Table 3: Subject-type and crossover values (Rand), by social distance……….28 Table 4: Quantile regression analysis – subject-type and crossover values (Rand)…...31 Table 5: Mean and median k-values and Area Under the Curve (AUC), by
subject-type………..31 Table 6A: OLS and Quantile regression analysis – subject-type and AUC………..32 Table 6B: OLS and Quantile regression analysis – subject-type and social discounting
rate………32 Table 7: Descriptive characteristics of recipients, by subject-type………35 Table 8A: Subject-type and recipient characteristics, by social distance –
Full sample...36 Table 8B: Subject-type and recipient characteristics, by social distance –
Sub-sample...39 Table 9A: Subject-type and relationship choice, by social distance – Full sample……..41 Table 9B: Subject-type and relationship choice, by social distance – Sub-sample……..42
List of Figures
Figure 1A: Crossover values (Rand), by subject-type – Full sample……….25 Figure 1B: Crossover values (Rand), by subject-type – Sub-sample……….26 Figure 2A: Subject-type difference in crossover values, by social distance – Full
sample………..29 Figure 2B: Subject-type difference in crossover values, by social distance –
13 Abstract
The social discounting task (SDT) is an experimental tool widely used by researchers to measure social discounting. Comparative social discounting studies employing the SDT amongst both laboratory and field subjects are limited. The current study compares giving and social discounting amongst laboratory and field subjects at a South African university, using an incentivized SDT. The study finds that for both subject types, crossover values declined as social distance increased. However, field and laboratory subjects differ on two of the three measures of altruism, namely the crossover value and the social discounting rate (k'), but no statistical difference with respect to AUC (Area Under the Curve). Furthermore, the study attempts to investigate the dynamics underlying differences in altruistic behaviour amongst field and laboratory subjects by examining the differences in the characteristics of the recipients selected at each social distance for laboratory as opposed to field subjects. The study employed both descriptive statistical analysis and regression analysis models. Other comparative studies have not collected detailed information on recipient characteristics and as such, this component of this research represents a novel contribution. The two subject groups selected very different recipients, which imply that the causes and consequences of altruism are different among the two groups of subjects participating in this experiment.
14
1. Introduction
Altruism may depend directly on perceived social distance. The closer one feels to someone else, the more altruistic one would be towards him/her. Although in general most humans behave pro-socially, i.e. taking other people’s interests into account, people are not generous to everyone alike and generosity declines as a function of social distance (Kurzban & Burton-Chellew, 2015; Rand & Nowak, 2013). Experiments undertaken by Jones and Rachlin (2006 & 2009) and Rachlin and Jones (2008) form the basis for the establishment of a scale by which social discounting, a situation where people assign more value to the welfare of close affiliates than they do to the welfare of distant affiliates (Sharp et al., 2012) could be measured. One particular experimental instrument used by these authors for the purpose of measuring this kind of pro-sociality is the social discounting task (SDT).
To date, most of the economic experiments employing social discounting tasks used laboratory subjects (students) as their main participants and these experiments can be classified as conventional based on the taxonomy of experimental design (Harrison & List, 2004). Recently Romanowich and Igaki (2017), for example, employed the SDT to test the effect of reward magnitude, alcohol and cigarette use on social discounting amongst 569 US and Japanese university students. Yi et al. (2016) used 50 undergraduate university students to complete the SDT while engaging in episodic future thinking about themselves and others, while Ma et al. (2015) measured differences in social distance-dependant pro-social behaviour amongst rural- and urban-reared students in China. Margittai et al.’s (2015) study investigated how a stressful experience influenced social discounting immediately after the experience, and within an hour after the stressful experience. Participants in the study were male only, German-speaking students, not enrolled for either Psychology or Economics study programmes from the Heinrich Hein University Düsseldorf. Strombach et al. (2013) employed the SDT with 206 students from Germany and China in order to measure the effect of cultural differences on generosity; and Ziegler and Tunney (2012) employed the SDT, together with a delay discounting task with 70 undergraduate students from the universities of Nottingham and Lincoln to determine whether there is an association between the delay choices people make for themselves and those made for others at different social distances. Jones and Rachlin (2006; 2008a; 2009) and Rachlin and Jones (2008b) used undergraduate students from the USA in all their experiments, while Osinski (2009) conducted the SDT experiment with 200 full-time students from the USA and Japan.
15
Experiments that are limited to a particular subject pool, and which make generalisations from their results about other groups in other social contexts are questionable (Carpenter et al., 2004). A common criticism of the relevance of conclusions drawn from laboratory experiments is that one needs to undertake an experiment with “real” people and not just with students (Harrison & List, 2004). The argument is proposed that if the researcher believes that students are not representative and expects different results from a different set of subjects, then the experiment should be conducted with a different set of subjects. Subjects’ responses and behaviour may differ due to the real-life context in which the experiment is conducted; differences in the socio-demographics; their social and cultural beliefs; and their relationship to the experimenter or other participants (Carpenter et al., 2004). There is an increasing number of other social discounting studies though in which field subjects are recruited in framed field experiments. Pornpattananangkul et al. (2017) used Strombach et al.’s (2014) research design and employed the SDT amongst 39 older adults and 39 younger adults from Singapore in order to measure the impact of age on social discounting. Sharp et al. (2012) applied the social discounting paradigm to boys who were 2nd to 12th graders recruited through community
organisations to investigate the relations between social discounting, age and externalising behavioural problems. Bradstreet et al. (2012) chose pregnant women to analyse social discounting amongst smokers, non-smokers and quitters. Yi et al. (2011) assessed delay, probability and social discounting amongst active methamphetamine users versus non-drug users.
However, comparative social discounting studies employing the SDT amongst both laboratory and field subjects are limited. The most recent study by Tiokhin et al. (2019) employed an adapted SDT amongst semi-literate, rural subjects from Bangladesh and Indonesia and US college undergraduates in order to test the validity versus generalizability of theories on human behaviour. The participants in Boyer’s et al. (2012) experiment were urban dwellers in Yueyang (China), ranging from students to employees, middle-class college students from the US as well as Kenyan herders. The study measured differences in time discounting; social discounting and generalised social trust amongst the subjects who were from different cultural backgrounds and found a highly significant effect of the location of the subjects on social discounting. This paper builds on this limited literature and involves the employment of an incentivised social discounting task among both field and laboratory subjects. This paper sets out by describing the altruistic behaviour of laboratory versus field subjects in South Africa as measured by giving and social discounting. Furthermore, the study attempts to investigate the
16
dynamics underlying differences in altruistic behaviour amongst field and laboratory subjects by examining the differences in the characteristics of the recipients selected at each social distance by laboratory, as opposed to field subjects.
2. Experiment
Participants:
The subjects were recruited over a period from August 2016 to September 2017 and June 2018. One hundred and eighty five students (laboratory) and 67 staff (field) subjects were recruited separately using flyers distributed among students and staff from the Faculty of Economic and Management Sciences at the University of the Free State. Two (2) experiments consisting of 10 sessions (6 staff and 4 student sessions) were conducted throughout the duration of the study.
Ethics:
Ethical clearance for the study was obtained from the Faculty of Economic and Management Sciences’ Research Ethics Committee at the University of the Free State (UFS-HSD2016/0124 and UFS-HSD2016/1084). Participation was voluntary and written informed consent was obtained from all subjects.
Procedure:
The recruitment flyers were physically distributed amongst undergraduate students during an Economics lecture at the University of the Free State and by hand to staff working in the Faculty of Economic and Management Sciences. A total number of 260 recruitment flyers were distributed amongst the students and 150 for staff. The staff and student experimental sessions were conducted separately on different dates and in different venues on the campus of the University of the Free State. Upon arrival at the venue, subjects were asked to present the recruitment flyer to the experimenter following which they were paid a show-up fee of R50. All subjects were required to read and sign a consent form before the actual experiment commenced. The experimenter provided each subject with a pencil, eraser and a hard-copy of the full experimental tool document. In all of the experimental sessions the subjects completed an incentivised pencil-and-paper version of Rachlin and Jones’ (2008b) standard social
17
discounting task (SDT). The experimenter read a set of detailed instructions and asked subjects to complete a practice table (Annexure A). Upon completion of the practice table, the experimenter instructed subjects to complete a series of seven uniform tables, one for each social distance. In each table, subjects made a choice between an amount of money for themselves versus an amount of money for the person cited at that specific social distance (Annexure B). The following instruction was included before the first table in the SDT in order to guide subjects in defining their social ladder:
Imagine you made a list of the 100 people closest to you in the world ranging from your dearest friend or relative at #1 to a mere acquaintance at #100. Now imagine the following choices between an amount of money for yourself and an amount for the #1 person on the list. Circle A or B on the right-hand side to indicate which you would choose in EACH line.
In each case, the #1 in the above instruction and in the table and other forms, including the recipient questionnaire, was replaced by #2, #5, #10, #20, #50, #100, respectively. The task was counter-balanced, with the tables in half of the experimental packages organised in the standard ascending order and in the other half in descending order. After completing the SDT each participant was also required to provide information on their social relation to the person, how often they communicated with the person, whether they lived with the person and, if not, how far they lived from the person, how often they visited the person, how long they had known the recipient, and how close they perceived themselves to be to this person on an emotional and psychological level, as well as the basic socio-demographics of this person (i.e. recipient characteristics, including gender, age, and household poverty rank of the person) (Annexure C). The questionnaire on the recipient characteristics immediately succeeded the respective social discounting table for that particular social distance. In other words, for each one of the seven SDT tables there was a corresponding recipient characteristics questionnaire. Subjects also completed a post-experimental questionnaire. The information on subjects collected by means of this questionnaire includes standard socio-demographic information: age; sex; race/ethnicity; education; household poverty rank; and individual financial status (Annexure D).
18 Payment:
Subjects received a show-up fee of R50. In addition, subjects earned some money from completing the experimental tasks. The payment protocol was as follows. Since the SDT was combined with other experimental tasks across differential experimental sessions, the specific task for actual payment had to be selected by rolling a ten-sided dice where 1-3 was for the Time Task, 4-6 was for Risk Task and 7-9 was for the SDT in the first experiment. An eight-sided dice was used for the second experiment, where 1-2 was for the Trust Game; 3-4 for the Ultimatum Game, and 5-6 for the SDT. Where the SDT was selected for payment, a seven-sided dice was used to select one of the social distances randomly. The subject would then proceed to roll a ten-sided dice to select one of the rows in the table and the experimenter would then record the decision taken by the subject (Option A or Option B) in that particular row, and proceed to implement the relevant decision for payment. If Option B, i.e. payment to the recipient at that particular social distance, was selected, the subject was required to provide the name, address and cell-phone number of the intended recipient. Payment of the monetary award to the recipient took place via a cashless money transfer payment to the cell-phone number of the recipient. The recipient would then be able to withdraw the money from an ATM of the relevant bank. Alternatively, if Option A, i.e. the subject choosing to keep the money for themselves, was selected, the monetary reward was transferred to the cell-phone number of the subject. The payment procedure was conducted individually and in private with the assistance of the experimenter (Annexure E). The mean earnings earned by subjects in these experiments amounted to R171.59.
3. Measures
This paper uses the Social Discounting Task (SDT) to measure altruism among student (laboratory) and staff (field) subjects. Three measures of altruism are estimated. A monetary value for the crossover value is estimated as the mean point at which a subject switched from choosing option A (selfish option) to option B (sharing option) for each social distance – and is captured as such in the task-level data set (i.e. for each social distance table). For instance, if in Row 1 of the table, a subject chose the selfish option at R180 instead of the sharing option of R160 for the recipient and then switched to the sharing option at R160 in Row 2, the crossover value was calculated to be R170. For subjects who exclusively chose the selfish option throughout, the crossover value is given as R0 and for subjects who exclusively chose the sharing option, the crossover value is assumed to be R190. For subjects who crossed over
19
multiple times, the crossover value was calculated at the first switch. The crossover value is employed as a measure of giving.
The crossover values are then used to determine social discounting functions which are fitted individually on each subject’s crossover values with the aid of Reed, Kaplan and Brewer’s (2012) Excel solver toolkit to derive a social discounting rate (k') for each subject. The social discounting function is assumed to be hyperbolic and following Jones and Rachlin (2006), it assumes the following form:
where νij represents the value that person i attaches to the welfare of person j, Ai represents the
value person i associates with her own welfare, and Nij is the rank person i assigns to person j
among i’s full list of associated people. The constant k′ , one aggregate measure of altruism, measures the steepness of discounting and the greater k′ is, the greater the degree of social discounting and the lower the degree of altruism (Sharp et al., 2012).
Another aggregate measure of altruism, i.e. Area under the curve (AUC) is derived from each participant’s series of crossover values. The AUC index is a normalised measure not dependent on functional form (Myerson et al., 2001) and is constructed for each subject by making use of the Excel solver toolkit of Reed, Kaplan and Brewer (2012). The value of this index varies from 1.0 (no discounting) to 0.0 (complete discounting) (Locey et al., 2011).
The use of these multiple measures of altruism, i.e. crossover value, k’, and AUC allows for robustness testing in the actual analysis. Two data sets were used, one using crossover values captured for each social distance (hence task-level data), and the other using aggregate measures of altruism, i.e. AUC and k', for each subject (hence subject-level data). For cross-over values, task-level analysis is carried out and subject- level analysis for AUC and social discounting rates. In both data sets subjects with inconsistent preferences on the SDT are identified in the study by distinguishing between subjects who switched multiple times and subjects who switched once or less per task. For the total number of 1 763 tasks completed by the 252 subjects, subjects switched multiple times in 200 or 11.34% of the total tasks while 43 subjects (17.06%) switched multiple times in at least one task. Multiple switching is very common in the SDT and Booysen et al. (2018) reported 41.1% and Sharp et al. (2012) reported 66.5% of such. For robustness checks, the analysis is split into a full sample analysis, which
20
includes all subjects, and a sub-sample analysis, where subjects who switched multiple times are excluded from the sample.
4. Analysis
A descriptive analysis of the socio-demographic information, i.e. age, race, gender, financial situation and language, of the subject pool was undertaken first separately for the laboratory and field subject groups and then for the combined sample. The study further employed descriptive statistical analysis to investigate whether laboratory subjects are more altruistic or less altruistic than field subjects by comparing the means and medians of the measures of altruism: crossover values, k-values and AUC. Chi-squared, t-tests and F-tests are used in the descriptive analysis. All three measures of altruism (i.e. crossover values, AUC and the social discounting rate [k'] are employed as dependent variables in the regression models estimating the relationship between these outcomes and subject-type (the independent variable of primary interest). These regressions are estimated for the full and sub-samples as well as at subject-level and task-subject-level (i.e. for each social distance task). Given the panel nature of the task-subject-level data (i.e. subjects are observed at multiple times in seven different choice sets, one for each social distance) a linear random effects (RE) regression model is employed to assess differences across subject-type. Subject-type is also interacted with social distances to determine if differences across subject-type are restricted to specific social distances.
The Shapiro-Wilk test was used to test for normality in all three outcomes, crossover; k' and AUC. In each instance, the statistical test and p-values confirmed the non-normality of the data: crossover (0.979, p<0.001); AUC (0.957, p<0.001); k' (0.247, p<0.001). As a result, quantile regression models were employed, since this type of analysis makes no assumptions on the normal distribution of the dependent variables. Quantile regression models are estimated using the crossover value, AUC index and k' as dependent variables and subject-type as the independent variable which is of primary interest.
A summary of the different regression models employed in the analysis is given below: Stage 1: Testing for differences in measures of altruism amongst laboratory and field
subjects
Regression: Random Effects Regression Model Dependent variable: Crossover value
Independent variables: social distance; subject-type; number of times switched; subject characteristics; family status; solidarity
21
Regression: Ordinary Least Squares Regression Models Dependent variables: (1) AUC and (2) k-value
Independent variables: subject-type; number of times switched; subject characteristics Regression: Quantile Regression Model
Dependent variable: Crossover value
Independent variables: social distance; subject-type; number of times switched; subject characteristics; family status; solidarity
Regression: Quantile Regression Models Dependent variables: (1) AUC and (2) k-value
Independent variables: subject-type; subject characteristics; number of times switched Subject characteristics include age; gender; household poverty rank; previous experimental experience.
Stage 2: Testing whether laboratory and field subjects select different recipients Regression: Ordered probit regression models
Dependent variables: (1) frequency of communication; (2) physical distance; (3) years known
Independent variables: social distance; subject-type; subject characteristics; number of times switched
Regression: Ordinary Least Squares Regression Models
Dependent variables: (1) psychological distance; (2) solidarity; (3) recipient age Independent variables: social distance; subject-type; subject characteristics; number of times switched
Regression: Probit Regression Model Dependent variable: recipient gender
Independent variables: social distance; subject-type; subject characteristics; household characteristics; number of times switched
Regression: Probit Regression Models
Dependent variables: Relationship – (1) family versus friend; (2) family versus other; (3) friend versus other; (4) family versus non-family
Independent variables: social distance; subject-type; subject characteristics; number of switches
22
For the crossover value analysis, the seven social distances are collapsed into four categories since the results of the regressions are not estimable across all social distances due to small sub-samples. The social distances were grouped as follows: category 1 – social distance #1 and #2; category 2 – social distance #5 and #10; category 3 – social distance #20 and category 4 – social distance #50 and #100. The solidarity variable was generated by applying the same approach followed by Booysen et al. (2018). The solidarity variable makes use of the Multiple Correspondence Analysis (MCA) from Bengston and Roberts (1991) and groups three recipient characteristics namely length of years known; physical proximity; and frequency of communication in order to provide an index against which relationship closeness can be measured.
The analysis therefore comprises two key stages. The first stage investigates differences in the levels of altruism amongst field and laboratory subjects, using each of the three measures of altruism. The second stage assesses whether field and laboratory subjects chose recipients with different characteristics, i.e. whether the underlying dynamics of altruism play out differently in the two groups of subjects. These analyses were completed for the full sample and the sub-sample, which excluded subjects who switched multiple times in the SDT, thus to determine the robustness of the results.
5. Results
The results of the study are presented in this section. First, the focus is on the comparison of subject characteristics between laboratory and field subjects. Then, the focus shifts to the comparison of giving and social discounting in laboratory and field subjects. Finally, the emphasis falls on the differences in the characteristics of recipients selected by laboratory and field subjects.
5.1 Subject characteristics
In Table 1, the characteristics of the subjects are reported separately for laboratory and field subjects and collectively for the total sample. The total sample consisted of 252 subjects, of which 185 were laboratory subjects and 67 field subjects. The median age is 23 years for laboratory subjects, 36 years for field subjects, and 25 years for the total sample, while the mean age is reported as 23.45 years for laboratory subjects, 36.60 years for field subjects, and 26.94 years for the total sample. The majority of the student subject group is Sesotho-speaking African females who are pursuing their studies within the Faculty of Economic and
23
Management Sciences, while the majority of the field subjects are Afrikaans-speaking white females who are employed within the Faculty of Economic and Management Sciences. Of the total sample, 93.42% of laboratory subjects are from the African race group, compared to only 37.88% of field subjects. Nearly 60% of the laboratory subjects speak Sesotho, compared to only 14.93% of staff. Both groups of subjects can be considered to be not poor with nearly 80% placing themselves on the 3rd and 4th rung of the Household poverty ladder. However, only
around 30% of subjects indicated they were not broke. Furthermore, 55.73% of laboratory subjects were considered broke and very broke, compared to only 25.38% of field subjects. A total of 91 subjects (36.4%) of the full sample indicated that they had previously participated in a similar experiment. Most of the differences in the characteristics of the subject groups are highly significant, particularly with respect to age, race, language and financial situation. This degree of imbalance between the two subject groups is expected given the nature of the type of subjects. Only gender and previous experimental experience showed no statistically significant differences across the two subject groups. The subject pools, therefore, are relatively diverse. One would therefore expect that giving and social discounting may differ significantly between laboratory and field subjects. The analysis now turns to this question.
24
Table 1. Sample descriptive statistics
Laboratory Field Total p-value
Age: Mean Median (IQR) 23 (20-26) 23.45 36(28-43) 36.60 25(21-29) 26.94 <0.001 <0.001 Female (%) 65.57 75.38 68.15 0.145 Race: African Coloured Asian/Indian White Total 93.41 1.65 1.65 3.30 37.88 7.58 3.03 51.52 78.6 3.2 2.02 16.13 <0.001 100.00 100.00 100.00 Language: Sesotho Afrikaans Setswana isiXhosa English isiZulu Other Total 59.89 3.30 9.89 7.69 3.85 6.04 8.80 14.93 5.97 47.76 5.97 7.46 16.42 2.98 47.79 15.26 8.84 7.63 7.23 5.62 7.22 <0.001 100.00 100.00 100.00 Faculty:
Economics & Management Natural & Agricultural Humanities Education Health Total 88.42 9.39 1.10 1.10 0.00 95.52 0.00 2.99 0.00 1.49 90.32 6.85 1.61 0.81 0.40 0.026 100.00 100.00 100.00
Household poverty rank: 1 2 3 4 5 6 Total 6.04 10.99 51.65 27.47 3.30 0.55 0.00 7.46 37.31 41.79 8.96 4.48 4.42 10.04 47.79 31.33 4.82 1.61 0.003 100.00 100.00 100.00 Financial situation: Very broke Broke Neither In good shape In very good shape Total 16.39 39.34 26.78 15.30 2.19 8.96 16.42 31.34 38.81 4.48 14.40 33.20 28.00 21.60 2.80 <0.001 100.00 100.00 100.00 Previous experimental experience (% yes) 38.25 31.34 36.40 0.315 Sample (n) 185 67 252
25
5.2 Giving and social discounting by subject-type
Figures 1A and 1B illustrate the distribution of the crossover values for the two subject pools for both the full sample and sub-sample. A greater number of field subjects crossed over at the higher crossover values compared to the laboratory subjects. At the other end of the spectrum, the opposite is observed, where more laboratory subjects crossed over at lower crossover values, compared to field subjects. A similar result is reflected in the sub-sample analysis reported in Figure 1B.
Figure 1A: Crossover values (Rand), by subject-type: Full sample
0 .00 5 .01 .01 5 .02 0 50 100 150 200 0 50 100 150 200 student staff De ns ity
Cross-over value (Rand)
26
Figure 1B: Crossover values (Rand), by subject-type: Sub-sample
Table 2A: Mean crossover values (Rand), by social distance and subject-type Social
distance Laboratory Field Full Sample p- Sub-Sample
value Total Laboratory Field value p- Total
1 128.43 147.01 0.005 133.37 120.66 145.08 0.008 127.68 2 121.08 141.49 0.004 126.50 114.11 138.16 0.001 120.73 5 107.89 123.43 0.032 112.02 98.79 116.10 0.020 103.50 10 94.00 101.49 0.323 95.99 86.72 97.50 0.163 89.60 20 80.27 91.34 0.142 83.21 71,67 87.30 0.028 75.95 50 73.04 75.07 0.795 73.58 64.97 68.36 0.650 65.87 100 64.05 64.02 0.997 64.04 56.16 55.57 0.937 56.00 Total 95.55 106.26 <0.001 98.40 86.68 100.94 <0.001 90.56 Sample (n) 1,294 469 1,763 1,138 425 1,563
Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches.
Table 2A shows the mean crossover values for the laboratory and field subjects. The mean crossover value for the full sample is R98.40, compared to R106.26 for staff subjects and R95.55 for student subjects. This difference between laboratory and field subjects is statistically significant (p<0.001). The mean crossover values for the sub-sample is lower than those of the full sample, with R90.56 reported for the total and R100.94 for staff subjects and
0 .00 5 .01 .01 5 .02 0 50 100 150 200 0 50 100 150 200 student staff D en sit y sd_crossover Graphs by Staff_Student
27
a mean crossover value of R86.68 reported for students. This difference is also statistically significant for the laboratory and field subjects (p<0.001). At each social distance, excluding the last distance of #100, the mean crossover value for the staff subject group is higher than that of the student subject group. However, these differences are only statistically significant at the close social distances (#1, #2 and #5, all at p<0.05) and are not statistically significant for the remainder of the social distances. In both groups, the majority of subjects crossed over at higher values for close social distances and at lower values for the more distant social distances. In the sub-sample, i.e. the group who did not switch multiple times in the experiment, the mean crossover values are lower than those of the full sample at each social distance. This could be attributed to the fact that subjects who did switch multiple times generally switched early on in the task, and therefore the mean crossover value for the sample inclusive of these subjects will be higher. For the sub-sample, the mean crossover value for staff is higher at all social distances, except for the last social distance, #100, where it is marginally lower. The differences in the means for laboratory and field subjects are statistically significant at social distances #1, #2, #5 and #20 (p<0.05).
Table 2B: Median crossover values (Rand), by social distance and subject-type Social
distance Laboratory Field Full sample p- Sub-sample
value Total Laboratory Field p-value Total
1 150 150 0.111 150 130 150 0.017 150 2 130 150 0.001 130 110 150 0.001 130 5 110 130 0.162 110 90 110 0.107 90 10 90 90 0.434 90 90 90 0.201 90 20 70 90 0.250 70 70 90 0.155 70 50 70 70 0.751 70 50 70 0.697 50 100 50 50 0.901 50 50 30 0.772 50 Total 90 110 0.001 90 90 <0.001 Sample (n) 1,294 469 1,763 1,138 425 1,563
Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches.
The difference between the full sample median crossover value for laboratory subjects (R90) and the field subjects (R110) is statistically significant (p<0.01). Furthermore, the differences in the median crossovers for the two subject types are only statistically significant for both the full sample and sub-sample at social distance #2, and for only the sub-sample at distance #1. The median crossover value for field subjects is higher than those of laboratory subjects in the full sample for three of the seven social distances (#2; #5; #20) and for the total result, although these results are only statistically significant for distance #2 and the total result. For the
sub-28
sample, the median crossover value of field subjects is higher than that of laboratory subjects for six of the seven social distances (#1; #2; #5; #20 and #50), but this result is only significant at distances #1 and #2. Both the mean and median crossover values for the full sample and sub-sample and for the two subject types decline as social distance increases.
Table 3: Subject and crossover values (Rand), by social distance
Full sample Sub-sample Model A:
Laboratory versus Field -1.932
(7.160) (7.766) -0.471 Wald chi2 (p-value)
R2 595.33 (<0.001) 0.329 367.19 (<0.001) 0.244
Sample (n) 1,709 1,516
LR-test (p-value) 1375.96 (<0.001) 1333.26 (<0.001)
Model B:
Laboratory versus Field -1.890
(7.153) (7.804) -0.507 Social distance (interaction)
1 9.533 (8.265) (8.918) 13.526 2 7.282 (8.556) (9.622) 8.737 5 3.931 (8.073) (8.839) 3.966 10 -6.912 (9.064) (9.741) -2.995 20 -2.051 (8.037) (8.456) -1.079 50 -11.468 (8.695) -10.953 (9.186) 100 -13.957 (8.702) -13.289 (9.344) Wald chi2 (p-value)
R2 639.83 (<0.001) 0.334 411.73 (<0.001) 0.249
Sample (n) 1,709 1,516
LR-test (p-value) 1398.69 (<0.001) 1355.12 (<0.001) Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches. In model A, a Random Effects (RE) regression model was applied. In model B, a RE regression model is applied with interactions of social distance and subject-type. Controls for age, gender, poverty status and experimental experience of sender.
Standard errors are reported in parentheses. Statistical significance: 1% ***, 5% **, 10% *
The Random Effects (RE) regression results are reported in Table 3. These regressions were estimated for both the full sample and the sub-sample. In model B, the effect of the subject- type, i.e. staff-student, on the crossover value at each social distance is estimated. When controlling for the differences in the characteristics of the subject pool, there is no difference in the level of altruism between laboratory and field subjects, not overall (model A) nor at each
29
of the different social distances (model B). At the closer social distances, 1, 2 and, 5 field subjects give more than laboratory subjects. At social distance 1, field subjects give R9.53 more than laboratory subjects, at social distance 2 field subjects give R7.28 more than laboratory subjects, and at distance 5 field subjects give R3.93 more than laboratory subjects. For the further social distances, i.e. distances 10-100, staff give less than students as indicated by the negative coefficients for distances 10, 20, 50 and 100. These results are mirrored for the sub-sample and are illustrated graphically in Figures 2A and 2B, with the most notable difference between the crossover values for the full sample of R6.91 compared to a R2.99 for the sub-samples at social distance 10. However, none of the differences reported here are statistically significant, suggesting some degree of balance in giving between the two subject pools.
Note: Comparison is for field versus laboratory subjects.
Figure 2A: Subject-type difference in crossover values, by social distance – Full sample
-3 0 -2 0 -1 0 0 10 20 D iff ere nce s in cro sso ve r va lu es (R an d) 12 5 10 20 50 100 SD_distance
30 Note: Comparison is for field versus laboratory subjects.
Figure 2B: Subject-type difference in crossover values, by social distance – Sub-sample
In Table 4, the results of the quantile regression models are reported. Three quantile regression models were estimated, one at the median (q = 0.50) and one each at the two quartiles (q = 0.25 and q = 0.75). The coefficients for the crossover values at the three quartiles are reported in Table 4 for the full and sub-samples, respectively. For the full sample, there is a strong, positive and statistically significant effect of the subject-type on crossover values for q = 0.25, which means that amongst the less altruistic subjects (q = 0.25) field subjects are willing to forego larger amounts than laboratory subjects. This result is significant at 1%, as opposed to 5% for the sub-sample. For the more altruistic subjects (q = 0.75), field subjects are willing to give less than laboratory subjects. This result is statistically significant for the full sample only, although the sign of the coefficient is also negative for the sub-sample. For subjects at the median level of altruism, i.e. q = 0.50, field subjects are willing to give less than laboratory subjects for both the full and sub-samples, although these results are not significant.
-4 0 -2 0 0 20 40 D iff ere nce s in cro sso ve r va lu es (R an d) 12 5 10 20 50 100 SD_distance
31
Table 4: Quantile regression analysis – subject-type and crossover values (Rand) Laboratory versus
Field (Full sample)
Pseudo R2 Laboratory versus
Field (Sub-sample) Pseudo R2 q = 0.25 8.352** (4.087) 0.223 10.520*** (2.329) 0.163 q = 0.50 -4.174 (4.495) 0.206 (4.599) -2.100 0.141 q = 0.75 -8.187** (3.568) 0,178 (3.096) -4.387 0.170 Sample (n) 1,709 1,516
Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches. Controls for age, gender, poverty status and experimental experience of sender. Standard errors are reported in parentheses. Statistical significance is reported as follows: 1% ***, 5% **, 10% *.
Table 5: Mean and median k-values and Area Under the Curve (AUC), by subject-type
Full Sample Sub-sample
Laboratory Field p
-value Total Laboratory Field pvalue - Total
AUC: Mean Median 0.401 0.377 0.419 0.362 0.630 0.887 0.367 0.406 0.360 0.344 0.377 0.326 0.638 0.565 0.365 0.337 Sample (n) 185 67 252 151 58 209 k-value: Mean Median 0.597 0.101 0.320 0.071 0.317 0.318 0.095 0.523 0.714 0.142 0.367 0.096 0.289 0.133 0.618 0.120 Sample (n) 185 67 252 151 58 209
Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches.
The differences in the mean and median crossover values reported in Tables 2A and 2B are also consistent with the mean and median results for the aggregate measures of altruism,
k-value and AUC as reported in Table 5. The mean k-k-value is lower for field subjects (0.320)
than laboratory subjects (0.597) and AUC higher for field subjects (0.419) than laboratory subjects (0.401), although none of these differences is statistically significant. This result is the same for both the full and sub-samples, i.e. social discounting does not differ statistically significantly between laboratory and field subjects. The question, however, is if these results prevail when using a regression framework.
32
Table 6A: OLS and Quantile regression analysis – subject-type and (AUC)
Full Sample Sub-sample A. OLS Regression
Laboratory versus Field F-statistic (p-value) R2 -0.049 (0.042) <0.001 0.187 -0.045 (0.045) 0.009 0.082 Sample (n) 246 204 B. Quantile regression q = 0.25 Pseudo R2 -0.028 (0.060) 0.083 -0.015 (0.061) 0.038 q = 0.50 Pseudo R2 -0.096 (0.070) 0.115 -0.070 (0.074) 0.049 q = 0.75 Pseudo R2 -0.055 (0.076) 0.138 -0.092 (0.080) 0.068 Sample (n) 246 204
Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches. Controls for age, gender, poverty status and experimental experience of sender. Standard errors are reported in parentheses. Statistical significance is reported as follows: 1% ***, 5% **, 10% *
Table 6B: OLS and Quantile regression analysis – subject type and the social discounting rate Full Sample Sub-sample
A. OLS Regression
Laboratory versus Field F-statistic (p-value) R2 -0.478** (0.233) 0.007 0.030 -0.574** (0.273) 0.162 0.023 Sample (n) 246 204 B. Quantile regression q = 0.25 Pseudo R2 0.010 (0.017) 0.009 0.020 (0.028) 0.005 q = 0.50 Pseudo R2 0.006 (0.037) 0.019 -0.015 (0.057) 0.007 q = 0.75 Pseudo R2 -0.171 (0.116) 0.028 -0.239 (0.149) 0.017 Sample (n) 246 204
Note: The full sample includes all observations, while the sub-sample excludes subjects who reported multiple switches. Controls for age, gender, poverty status and experimental experience of sender. Standard errors are reported in parentheses. Statistical significance is reported as follows: 1%***, 5%**, 10%*
